You made the work. Nobody asked whether it could be fed into a machine that learns to imitate it — it just happened, quietly, at internet scale. If that sits wrong with you, you’re not powerless. There are real, concrete steps you can take today to tell AI companies: not my stuff.
Here’s the honest version of how opting out works in 2026, in plain English — what to click, where to go, and (just as important) what these tools genuinely can and can’t do. No law degree required.
First, the truth nobody puts on the button
Opting out of AI training is real and worth doing — but it is mostly voluntary on the other side. There is no master switch that yanks your work out of every AI model on Earth. What you’re actually doing is posting a clear, machine-readable “do not train” sign that cooperating companies have agreed to respect.
Track the cases yourself
Every current AI-copyright ruling — and the ones still being fought — lives in our AI Copyright Ruling Tracker. Filter by your creator type to see, in plain language, how each case affects your rights.
Two things follow from that. First, the well-behaved crawlers — the ones from major, brand-name AI companies — largely do honour these signals, because they don’t want the legal and PR headache of ignoring them. Second, none of this is retroactive: if your work was already scraped into a dataset two years ago, an opt-out today stops future training, not the past. Think of it as closing the gate going forward. That’s still valuable — most models get retrained and updated constantly. Let’s close the gate.
Step 1: Lock down your own website
If you have your own site, blog, or portfolio, this is the one place you have full control. AI crawlers are supposed to check a file called robots.txt at the root of your domain before they take anything. You can tell the training crawlers to stay out.
Add a robots.txt file (or edit the one you have) with lines like these:
User-agent: GPTBotDisallow: /User-agent: Google-ExtendedDisallow: /User-agent: CCBotDisallow: /User-agent: ClaudeBotDisallow: /
In plain terms: GPTBot is OpenAI’s training crawler, Google-Extended controls whether Google can use your content to train its Gemini models, CCBot is Common Crawl (a giant public dataset many AI models are built from), and ClaudeBot is Anthropic’s. Blocking these does not hurt your normal Google or Bing search rankings — regular search uses different crawlers. You’re only closing the AI-training door.
On WordPress, you can usually edit robots.txt through an SEO plugin. On other platforms, look for a “robots.txt” or “crawler” setting. If you’re on Cloudflare, it now offers a one-click option to block AI training bots, plus a newer “Content Signals” line in robots.txt that states how your content may be used after it’s fetched.
One caveat: robots.txt is the polite-request layer. Good bots obey it; bad bots ignore it. For the ones that ignore it, only server-level blocking (via Cloudflare, a firewall rule, or your host) actually stops them.
Step 2: Flip the opt-out switches on the platforms you post to
Most creators live on other people’s platforms — Instagram, YouTube, DeviantArt, Tumblr, LinkedIn. Each one has its own setting buried somewhere in your account, and they change often, so poke around “Settings > Privacy” or “Data” and search for “AI” or “generative.” A few current examples:
- Meta (Instagram / Facebook): Look for the “AI at Meta” or data-usage objection form under your account’s privacy settings. In some regions you can formally object to your public content being used for AI.
- DeviantArt & Tumblr: Both offer a “noai” / “no AI training” toggle that flags your content so cooperating crawlers skip it.
- LinkedIn: Under Settings > Data Privacy, there’s a toggle for using your data to improve generative AI — turn it off.
- Adobe: If you use Creative Cloud or the Content Authenticity app, you can set a “Generative AI training and usage” preference on your account.
These toggles only cover that platform, and only from the moment you flip them. But they’re free and take minutes, so work through the handful of places you actually post.
Step 3: Register with a Do Not Train registry
This is the closest thing to a central opt-out. Spawning runs Have I Been Trained and a global Do Not Train Registry — a shared list that participating AI companies check before scraping. You can search whether your images already appear in big public datasets, then add your work (or your whole domain) to the registry to reserve your rights.
The registry is huge and growing — tens of millions of registered works — and major players including Stability AI and Hugging Face have agreed to honour it. It isn’t universal law, but it’s the biggest coordinated opt-out that exists, and registering costs nothing. If you make visual art, this is one of the highest-leverage ten minutes you can spend.
Step 4: Attach “do not train” to the files themselves
Website settings and platform toggles protect a location. But your files travel — someone reposts your art, and the opt-out you set on your own site doesn’t come along for the ride. The fix is to bake the signal into the file itself using Content Credentials (the C2PA standard).
Content Credentials are like a tamper-evident nutrition label attached to a file: who made it, how, and — crucially — a “do not train” preference (technically the c2pa.ai_training assertion). Adobe’s free Content Authenticity web app lets you attach these credentials, including a signal that you don’t want the content used to train AI, and that preference is already recognised by Adobe Firefly and Spawning.
Be clear-eyed: Adobe itself calls this a “preference,” not a guarantee, and adoption across every AI company is still a work in progress. But it’s the direction the industry is moving, and it makes your intent travel with the work instead of staying pinned to one web address.
Step 5 (optional): Active protection tools
Everything above is a request. If you want something that fights back, visual artists have two free tools from the University of Chicago:
- Glaze subtly alters your image so AI models struggle to learn your personal style from it, while it still looks the same to human eyes.
- Nightshade goes further — it “poisons” the image so that models which train on it anyway learn the wrong things, discouraging scraping in the first place.
These aren’t magic and can be worked around, but they raise the cost of using your work without consent. For an artist whose style is the whole business, they’re worth a look.
What opting out can’t do — and a quick honest note
Let’s not oversell it. Opting out is forward-looking, not retroactive; it depends on companies choosing to cooperate; and it won’t reach every scraper on the internet. What it does do is establish, clearly and on the record, that you did not consent — which matters as laws, licensing deals, and court cases increasingly turn on exactly that question. Setting these signals now also positions you to benefit if future rules or settlements favour creators who reserved their rights.
So do the quick wins today: your robots.txt, your platform toggles, and a Do Not Train registration. Add Content Credentials and Glaze/Nightshade if your work warrants it. None of it takes long, and together it turns “I never agreed to this” from a feeling into a documented fact.
A light note: this is general information to help you understand your options, not legal advice. If real money or a specific dispute is on the line, talk to an IP attorney about your situation.
Sources & further reading:
- Spawning — Have I Been Trained / Do Not Train Registry
- Spawning Blog — Opt-outs that Work in the World of AI Data Scraping
- OpenAI — GPTBot and how to disallow it in robots.txt
- Google — Google-Extended and controlling AI training access
- Adobe — Content Credentials & Generative AI Training and Usage Preference